Voice memos to Markdown: a local-first workflow

Overview

Voice memos are fast to capture and slow to reuse if they stay audio-only. A local-first workflow converts recordings into Markdown—transcript, summary, and links you can search in Obsidian or any text editor.

This guide describes the pattern end to end: capture, sync, transcribe, summarize, and file notes. It is tool-agnostic where possible, with a short path to automating the same steps in Nucleate.

The basic loop

Record → (sync) → Transcribe → Summarize → Markdown in vault → Review

Each stage can be manual or automated. The goal is consistent folders and formats so notes compound over time instead of piling up as orphaned .m4a files.

1. Capture audio

Desktop: built-in Voice Recorder (Windows), QuickTime, or any app that exports .wav, .mp3, or .m4a.

Mobile: use a recorder that saves to a specific folder or shares into cloud storage. On Android, apps like Easy Voice Recorder work well with a sync folder; see Cloud & mobile for a concrete Android setup.

Tips:

• Short, frequent memos beat one long weekly dump for daily summaries
• Pauses and restarts are fine—local models tolerate imperfect speech
• Name files with dates if your recorder supports it (2026-06-13-walk.md later pairs with audio)

2. Get files where processing happens

Your transcription machine needs a local copy of new audio.

See Obsidian offline sync for sync tradeoffs.

3. Transcribe locally

Run a local Whisper-style model so audio does not upload to a third party. Options include:

• Nucleate — watches a folder and transcribes automatically
• Whisper CLI / faster-whisper — manual or scripted batch runs
• Other local apps — export text, then paste or save into the vault

For privacy and cost tradeoffs, read Local vs cloud transcription.

Formats: .wav works everywhere; .mp3 and .m4a typically need FFmpeg installed. Nucleate handles conversion when FFmpeg is present—see FFmpeg setup.

4. Summarize (optional but valuable)

Raw transcripts are searchable but long. A local LLM pass produces bullets, decisions, and next steps.

1. Install Ollama (or the macOS guide)
2. Summarize in the terminal for one-offs, or use Nucleate for structured daily/weekly output

Nucleate applies user modes and optional Insights on top of the transcript so notes match your workflow (developer log, journal, meeting review, etc.).

5. Write Markdown into Obsidian

A minimal daily note structure:

---
date: 2026-06-13
source: voice-memo
---

## Summary
- ...

## Transcript
...

Keep summaries in a folder you browse often and archives (full transcripts) one level deeper if you want a cleaner graph view.

Link new notes from an index page or use Obsidian’s daily notes plugin—consistency matters more than which plugin you pick.

Manual vs automated

Manual pipeline

1. Drop audio into a inbox/audio folder
2. Run transcription tool; save .md to notes/inbox
3. Run Ollama or edit summary by hand
4. Move file to notes/daily/YYYY-MM-DD.md

Automated pipeline (Nucleate)

1. Set Incoming audio to synced or local folder
2. Set output folders to vault subfolders
3. Enable automation; new files transcribe and summarize in the background
4. Review in Obsidian when ready

Nucleate also supports weekly and monthly rollups when you record across the week—see How it works.

Troubleshooting

Transcripts appear but summaries do not
Check Ollama is running and a model is pulled. See Software prerequisites.

Files never picked up
Confirm the recorder saves where sync and Nucleate expect. Paths must match Menu → File Sync → Folder settings after any folder rename.

Notes sync but audio does not (or vice versa)
Large audio files may sync slower than Markdown. Process on desktop first, then let Nucleate archive audio locally to keep the vault light.

Using this with Nucleate

Nucleate implements this workflow as a tray app: watch folder, transcribe, summarize, write Markdown, optional Notion push. It is free for core automation; Pro adds Labs and unlimited weekly processing.

Start here:

1. Installation
2. Cloud & mobile if phone capture matters
3. Talking out loud if recording feels awkward at first

Related guides

• Obsidian offline sync
• Local vs cloud transcription
• Ollama on macOS (Apple Silicon)